THE POLITICAL ECONOMY OF PATRONAGE: EXPENDITURE PATTERNS IN THE ARGENTINE PROVINCES,

Transcription

1 THE POLITICAL ECONOMY OF PATRONAGE: EXPENDITURE PATTERNS IN THE ARGENTINE PROVINCES, Karen L. Remmer Department of Political Science Duke University Box Durham, NC

2 ABSTRACT Under what conditions do politicians emphasize patronage allocations over the provision of public goods? Building upon research on democratic policy management, this paper aims to improve our understanding of patronage politics by focusing upon the political incentives influencing the ability and willingness of politicians to target public sector allocations to political supporters. Drawing upon data on spending priorities at the provincial level in post Argentina, the statistical analysis provides evidence that the relative importance of patronage allocations fluctuates with partisanship, electoral cycles, revenue sources, and public sector investment in economic development. The findings underline important and largely neglected parallels between clientelistic and programmatic politics and thereby have important implications for the study of the political economy of democracy. 1

3 The Political Economy of Patronage: Expenditure Patterns in the Argentine Provinces, Research on the political economy of democracy takes programmatic linkages between citizens and politicians as its central point of theoretical departure. Whether the focus of analysis is economic voting, fiscal policy choice, or political business cycles, the underlying presumption is that democratic accountability revolves around voter responsiveness to the programmatic appeals and policy performance of elected officials (Alesina, Roubini, and Cohen 1997; Alesina and Rosenthal, 1995; Franzeze 2002; Hibbs 1987; Persson and Tabellini 2000; Tufte 1978). For the majority of the world s voters, however, the electoral calculus is conditioned less by programmatic considerations than by patron-clientelism. The difference is pivotal for the study of policy choice and performance. Whereas programmatic exchanges establish motivations for the provision of collective goods, including macroeconomic growth and stability, patronage politics revolve around the targeting of public resources to political networks or machines. What sociopolitical conditions lead politicians to emphasize the provision of patronage over the allocation of public goods? How, if at all, do electoral cycles and partisanship affect the relative priority placed on patronage allocations? Given the welldocumented prevalence of patronage politics across the less industrialized world, as well as the evidence regarding its economic and political costs, such questions are of major importance. The construction of political reward networks around the distribution of public sector jobs and resources has been repeatedly linked with deficit spending, public sector inefficiency, resistance to market-oriented reform, macroeconomic instability, state predation, and reduced economic growth (Van de Walle 2001; Kitschelt and Wilkinson, n.d.; Keefer 2004; Keefer and Vlaicu 2004; Brinkerhoff and Goldsmith 2004; Chubb 1981).

4 No less important are the political consequences of patronage, which are seen to include politicized bureaucracies, corruption, electoral manipulation, ethnic voting, incohesive political parties, political inequality, the consolidation of incumbency advantage, fragmented civil societies, attenuated forms of citizenship, and fragile political institutions (Graziano 1976; Wantchekon 2003; Fox 1994; Hagopian 1996). The obvious implication is that privileging the provision of patronage over public goods has significant consequences for both economic development and democracy. Although the available literature offers many rich insights into patronage politics, we are far from being able to offer persuasive answers to basic questions about its relative prevalence or importance. The problems are threefold. First, previous research has tended to conceptualize patronage politics less as an instrumental and distributive exchange between politicians and their supporters than as a distasteful form of governance fundamentally antithetical to democratic rule (Diamond 1999: 81, 244; Fox 1994). Indeed, the concept of patron-clientelism has been used to characterize entire polities as undemocratic (e.g. Martz 1997), inhibiting analysis of the factors that induce politicians to adopt patronage strategies of governance. Second, most prior research has been conducted on the basis of case studies, making it difficult to assess the validity of theoretical claims or draw broader inferences about causal patterns. Third, the traditional answers to questions about the predominance of patronage allocations over the provision of public goods revolve around the social composition of the electorate, neglecting the role of politicians. In contrast, this paper places politicians and policy choice at the center of analysis and follows other recent research (e.g., Kitschelt 2000; Kitschelt and Wilkinson, n.d.; Hopkin 2001; Medina and Stokes 2002; Robinson and Verdier 2003) in treating patronage 2

5 politics as eminently compatible with democracy. The basic conception is that politicians have various ways of constructing a base of political support: by targeting or promising to target government resources to political supporters (patronage politics), by allocating or promising to allocate collective goods to broad sectors of the electorate (programmatic politics), or some combination of the two. The virtues of thinking about patronage politics in this way are numerous. As a start, the conception accommodates the otherwise awkward observation that patronage-based machines dominate electoral interactions in many polities that we classify as democratic. In addition, and of greater significance for the study of political economy, the conceptualization establishes the basis for bringing the literatures on patron-clientelism and democratic policy management together in ways that advance our understanding of both phenomema. Drawing both bodies of literature, this study attempts to shed new light on the factors leading politicians to privilege patronage over the provision of public goods. Central emphasis is placed upon the incentives for politicians to expand public sector employment to reward political supporters, co-opt opponents, and thereby advance their political careers. Drawing upon a statistical analysis of spending priorities at the provincial level in Argentina in the post-1983 period, the results emphasize that patronage strategies of governance cannot be understood merely as a reflexive response by politicians to the sociology of the electorate. Politicians shape the attractiveness and credibility of electoral appeals built around the distribution of patronage through their partisanship, policy choices, and responses to the electoral cycle. 3

6 Prior Research Working over a period of several decades, anthropologists, political scientists, historians, economists, and sociologists have all made significant contributions to the study of patronage politics (Graziano 1973, 1976; Eisenstadt and Roniger 1981, Lemarchand 1972, Powell 1970, Roniger and Gunes-Ayata 1994; Gay 1994; Scott 1969, 1972; Shefter 1994; Strickon and Greenfield 1972, Weingrod 1968, and Wolf 1966). The traditional focus of research has been social interactions at the micro-level; however, conceptions of patronage differ considerably, reflecting the disparate nature of the phenomena grouped together under a common rubric as well as disciplinary divides. Scholars specifically interested in political patronage have tended to downplay the importance of individual exchanges in favor of the study of political behavior at the meso- or macro-level of analysis. Thus whereas anthropologists focus on peasant-landlord types of personal relationships, for political scientists patronage politics is largely the study of how political party leaders seek to turn public institutions and public resources to their own ends (Weingrod 1968). These activities may involve individual or collective exchanges, but in either case they entail the targeting of public sector resources to specific members of society to the exclusion of others. As suggested by Keefer (2004), the implication is that attributing problems regarding the underprovision of public goods to patronage politics is largely tautological--by definition patronage politics privilege selective incentives over the delivery of public goods. Political scientists also differ from colleagues in related disciplines by placing the exchange of public sector employment for political support at the center of analysis (e.g., Gordin, 2002; Robinson and Verdier 2003; Calvo and Murillo 2004). As emphasized by Stokes (2005), the efficacy of clientelistic electoral appeals is less a product of the income, 4

7 education, or other characteristics of voters than of the capacity of politicians to monitor electoral behavior. The distribution of public jobs to supporters who gather information about voters and address their material needs is critical to this capacity. It is precisely for this reason that patronage politics is often utilized as a synonyms for patrimonialism, which in Weber s (1947) classic formulation represents a form of governance in which bureaucratic appointment revolves around personal connections, informal rules, and the blurring of the public and private realms. Emphasizing the informal, ad hoc, and potentially illegitimate chararacter of patronage politics, however, underlines a widely shared research dilemma. Efforts to document and systematically compare patronage politics run up against acute problems of data reliability and validity. Traditionally, researchers have attempted to address this problem by emphasizing in-depth field research and case study methodology. More recently, however, the study of patronage politics has drawn upon alternative approaches. First, researchers attempting to assess the relative significance of patronage politics have begun relying on public opinion surveys (e.g. Brusco, Nazareno, and Stokes (2002) or a combination of survey research and experimental methods (Wantchekon 2003). While the resulting body of evidence offers interesting findings regarding the relative prevalence of patronage in particular countries as well as the varying importance of material distributions to different kinds of voters, such data fail to capture political strategy or behavior at the elite or policy making level, where the key exchanges of public jobs and resources occur. The second major alternative to the tradition of in-depth field research involves comparative analysis of the provision of public employment or selective benefits. To assess changing patterns of patronage politics in Spain, for example, Hopkin (2001) studies 5

8 spending priorities and the targeting of public expenditures over time. Magaloni, Diaz- Cayeros, and Estevez (2002), Chhibber and Nooruddin (2004), and Calvo and Murillo (2004) engage in a parallel effort to explain policy variations at the subnational level. Building on this second tradition of examining the supply or public policy side of patronage politics, this paper focuses on explaining variations in patronage allocations with specific emphasis on spending on public sector employment. Theoretical Perspective Given the obvious inefficiencies, why do politicians resort to the distribution of public sector jobs and resources rather than build support on the basis of policy performance or programmatic commitments? The answer to this question has traditionally revolved around the issue of political demand, with the efficacy of patronage strategies seen as varying directly with the poverty, inequality, and economic uncertainty of the electorate (e.g., Migdal 1988; Auyero 2000; Brusco, Nazero, Stokes 2002; Calvo and Murillo 2004; Medina and Stokes 2002; Alesina, Baqir, and Easterly 1998). Focusing instead on the logic of political supply, the subsequent analysis is designed to explore factors shaping the incentives and opportunities for politicians to build support networks around patronage allocations. First, extending the logic regarding the importance of voters s poverty and economic insecurity, it can be anticipated that the supply of patronage will vary inversely with public investment in economic development and efficient forms of social redistribution. Incumbents generating support through patronage networks have less incentive to pursue socioeconomic development than their programmatic competitors or counterparts, helping to explain persisting inequalities and developmental failures in 6

9 countries dominated by patronage politics. The less public policy is oriented toward the provision of public goods, the greater the credibility and attractiveness of electoral appeals built around patronage networks. The incentives for politicians to supply patronage to their supporters will therefore vary with the level of other policy commitments. The logic of this line of analysis is developed formally by Medina and Stokes (2002) and Robinson and Verdier (2003) as well as documented empirically on the basis of research on Africa (Bates 1981) and Southern Italy (Chubb 1982). Second, and related to the foregoing argument that the supply of patronage is shaped by the level of other policy variables, the incentives to expand public employment and other forms of patronage distribution will vary with the sources of state income. The higher the level of non-tax revenues, including external aid and transfers, the greater the incentives for state expansion (Remmer 2004; Rodden 2002; Rodden and Wibbels 2002). The argument parallels the discussions of the resource curse, according to which high levels of dependence on oil or other revenues not generated by domestic taxpayers create opportunities for bureaucratic expansion that bias policy in the direction of patronage (Ross 2001). Third, building on research on the political economy of elections in advanced industrial societies (e.g., Hibbs 1987; Alesina, Roubini, and Cohen 1997), the supply of patronage rewards is likely to vary with partisanship. According to Calvo and Murillo (2004), the preference of parties for patronage spending varies with their relative success in cultivating constituencies that are susceptible to clientelistic appeals. Since the electoral gains from targeting benefits to supporters may be expected to rise as the economic security and income of voters declines, variations in the electoral base of parties are therefore 7

10 expected to be reflected in their propensity to engage in patronage politics. Fourth, although patronage has been treated as a relatively resilient or sticky property of political life, the incentives to distribute patronage are likely to fluctuate over time in accordance with the electoral cycle. Setting to one side the caveats introduced by rational expectations models, the conventional wisdom about democratic policy management emphasizes the reliance of voters upon retrospective judgments about the state of the economy, which creates incentives for politicians to pursue expansionary policies prior to elections. As suggested by Ames (1987:13), however, politicians whose political careers depend on building and maintaining an active political network of supporters face pressures to increase spending after elections. The key expectation regarding the distinctiveness of patronage business cycles is thus that distributions of patronage should expand significantly in the year subsequent to elections as politicians attempt to co-opt opponents, deliver on promises of jobs to supporters, and otherwise fortify their political bases to advance their political careers. Given the centrality of the exchange of public sector jobs for the mobilization of voters by party activists, what matters is less retrospective performance in the eyes of voters than the prospective commitments and ambitions of electoral winners. Drawing upon the preceding theoretical considerations, the statistical analysis presented in this paper addresses a set of relatively straightforward research hypotheses about variations in patronage allocations. The first hypothesis draws upon conventional explanations for variations in the provision of patronage by focusing on the issue of electoral demand. The remaining hypotheses are designed to explore an alternative, but potentially complementary, line of theoretical analysis that places central emphasis on the 8

11 strategic choice of incumbent politicians to supply patronage. H1: The level of resources allocated to patronage employment increases with the economic insecurity of the electorate. H2: The lower the level of public sector investment in economic growth, the greater the tendency to allocate resources to patronage. H3: The higher the reliance on non-tax revenues, the greater the tendency towards the expansion of patronage spending. H4: The allocation of resources to public sector employment reflects variations in partisan control. H5: Patronage allocations tend to expand in the wake of elections, giving rise to a distinctive patronage-driven political business cycle. Research Design Data for the analysis are drawn from the study of electoral outcomes and policy performance in the Argentine provinces since the restoration of democracy in The theoretical and methodological advantages of this subnational focus are several. First, relying on intra-national comparisons means that a number of variables that might influence spending on public employment are held relatively constant, including the political regime, culture, class linkages of major political parties, and important rules of the electoral game. Among the latter are balloting procedures, which affect the capacity of politicians to monitor the behavior of the electorate, even where voting is secret. Across Argentina, for example, electoral ballots are produced by political parties, enhancing the leverage of party bosses over voters as well as inhibiting split ticket voting. Given the paucity of prior research capable of addressing the relative importance of such electoral practices, intra- 9

12 national comparisons offer a promising basis for advancing research. Second, given the contribution of provincial fiscal policies to the largest sovereign default in history, the Argentine experience is also particularly interesting and important. Since 1983, provincial spending has far outstripped revenue growth, leading to mounting indebtedness and problems of economic adjustment at the national level (Jones, Sanguinetti, and Tommassi 2000, Remmer and Wibbels 2000; World Bank 1996). The crux of the difficulty has been increased personnel spending. Whereas capital spending at the provincial level declined from U.S. $4.6 billion to U.S. $2.1 billion over the course of the 1980s, World Bank estimates indicate that provincial public employment simultaneously increased almost 40 percent (World Bank 1993:129). The same trends persisted into the 1990s, with spending on provincial administration more than doubling between 1990 and 1995 alone. Taking the period as a whole, provincial personnel spending expanded at an annual rate of 7.6 percent ((Ministerio de Economía y Producción, a) more than four times the average rate of GDP growth (World Bank 2005). Despite the negative implications of these trends for national macroeconomic stability, efforts to cut provincial payrolls have run up against severe resistance, mainly due to the heavy reliance of provincial party machines on the distribution of government jobs. Hence despite two decades of stabilization effort, profligate provincial spending continued through the turn of the century, ultimately contributing to the Argentine economic meltdown of Third, as a result of the administrative decentralization that has occurred in Argentina over the past few decades, control over the provincial public sector has assumed major importance. Governors command considerable political power, enjoy high levels of 10

13 visibility on the national political stage, and allocate a significant proportion of government resources. According to the Inter-American development bank, Argentina is more decentralized than any other country in Latin America, with the provinces accounting for more than 90 percent of public spending on primary and secondary education, as well as more than 90 percent of spending on public housing, 74 percent on public health, 50 percent on public safety and justice, and 57 percent on infrastructure and services (Inter- American Development Bank 1994). In addition, the Argentine provinces hold regular elections for provincial executive and legislative office, making it possible to explore research hypotheses about such issues as partisan control and political business cycles at the subnational level of analysis. Fourth, Argentine parties have been relatively well institutionalized and have played a powerful role in structuring electoral competition, facilitating analysis of the impact of partisanship on patronage spending. Since the late 1940s, the two main partisan competitors have been the Peronist (or Justicialista) party and Radical party (or Unión Cívica Radical), whose combined strength averaged 70 to 88 percent of the gubernatorial vote between 1983 and Fifth, a rich base of prior research has documented the importance of patronage politics in Argentina. Variously drawing upon ethnographic research, aggregate data, and survey responses, scholars have emphasized the importance of patronage networks to interand intra-party competition, as well as the partisan bias introduced by federal transfers supporting public employment at the provincial level (Levitsky 2003; Gibson, Calvo, and Falleti 2004; Calvo and Murillo 2004; Auyero 2001; Powers 2001; Brusco, Nazareno, and Stokes 2002; Stokes 2005). 11

14 Finally, assessing the causal nexus between variables such as economic insecurity, partisanship, and electoral cycles, on the one hand, and patronage spending, on the other, requires data on a reasonably large and extended series of elections as well as consistent and reliable cross-sectional time-series data on fiscal and economic performance. Outside of the advanced industrial world, these requirements are not easily met. Given the availability of data at the provincial level for Argentina, practical considerations thus buttress the theoretical advantages of focusing on that case. The extent to which the findings based on the experience of a single nation establish the basis for broader theoretical generalization is always open to question. Nevertheless, in contrast to most prior research on patronage politics, which has either relied heavily on ethnographic research in particular regions, cities, or neighborhoods over a limited time period or drawn upon statistical evidence of dubious validity and cross-national comparability, intra-national comparisons allow for meaningful and systematic comparisons and thereby establish the basis for broader theoretical claims. The distinctive contributions of this study revolve around its capacity to marshal annual data covering a relatively extended period of time to analyze the political economy of patronage. Data and Methodology To explore the hypotheses outlined above, the subsequent empirical analysis draws upon an original data set on the Argentine provinces covering the period. The dependent variable is patronage spending, which is measured on the basis of provincial personnel allocations. Given the centrality of the provincial arena to the construction and maintenance of political careers, the lack of a civil service system, and the importance of provincial governments to local economies, control over appointments to provincial 12

15 administrations is a pivotal political resource in Argentina. Three specific indicators of personnel spending are utilized: total personnel spending in 1991 peso values, personnel spending per capita in 1991 peso values, and personnel spending as a percentage of total provincial spending net of interest payments on the provincial debt. All three measures are based on official government data (Ministerio de Economía y Producción, a). Although the absolute level of personnel spending is statistically correlated with the other two indicators at the.05 level of statistical significance, the three indicators capture differing dimensions of policy. Whereas the first reflects the absolute magnitude of patronage spending, the per capita and share indicators, respectively, measure spending relative to the potential level of patronage demand and the priority placed on patronage spending within the provincial public sector. It may be noted that total personnel spending is highly correlated with the available official data on the number of provincial employees (r=.959) (Ministerio del Interior ). The independent variables of theoretical interest are operationalized as follows. First, economic uncertainty is measured on the basis of provincial unemployment data as reported by the Argentine national statistical institute (INDEC ). Over the past two decades, unemployment has represented the key source of economic insecurity in Argentina, with rates for some years approximating 20 percent. Unlike other available indicators of economic performance, unemployment has also varied widely across the provinces, averaging a low of 3.9 percent in Santa Cruz to 14.0 percent in Tucuman, and has been measured on a regular basis. Drawing upon the federal government s NBI indicator (INDEC 2005), these data are supplemented with estimates of poverty levels, which are interpolated from the limited available government data on basic needs. 1 13

16 Second, the hypothesis regarding public sector investment in economic development is assessed on the basis of data on capital spending in accordance with the logic elaborated by Chhibber and Nooruddin (2004). While significant leakages may be associated with public sector investment (Tanzi and Davoodi 1997), capital spending yields public or at least quasi-public benefits rather than the more narrowly targeted gains associated with personnel spending. As noted above, democratization in Argentina was associated with a dramatic decline in provincial capital spending, although the pattern varied considerably from province to province. Third, variations in a province s relative dependence on tax and nontax revenues are assessed by three indicators drawing upon official data (Ministerio de Economía y Producción b): (1) tax revenues, which include national tax revenues transferred to the provinces according to a fixed formula as well as provincial tax revenues; (2) discretionary federal transfers, and (3) non-tax provincial revenues. While both of the latter represent non-tax revenues, they are disaggregated to assess potential differences in their impact and relative importance. Fourth, the impact of partisanship and electoral cycles is measured on the basis of four indicators. Drawing upon prior research emphasizing the linkage between Peronist governance and public sector patronage in Argentina (Calvo and Murillo, 2004), the impact of partisanship is measured on the basis of dummy variables for Peronist control of the presidency and the provincial governorship. Evidence of a patronage business cycle involving increased spending in postelectoral years is assessed on the basis of dummy variables for gubernatorial election years and their leads and lags. With 1983 representing the baseline (the year of transition from authoritarian rule to democracy), the analysis 14

17 covers five gubernatorial contests in Argentina's twenty-three provinces. 2 Finally, because prior research suggests that variations in the structure of political competition are likely to influence the supply of patronage (Bueno de Mesquita, Morrow, Siverson, and Smith 2000; Chhibber and Nooruddin 2004; O Dwyer 2004; Maglioni 2005), the analysis controls for electoral competitiveness as assessed on the basis of the percentage vote received by the winning candidate in the most recent gubernatorial election. In the analysis that follows, alternative measures of competition are also evaluated. For theoretical reasons, the statistical estimations rely upon an error-correction model similar to that utilized in other recent research on government spending (e.g., Iverson and Cusack 2000; Kaufman and Segurra, 2001; Rodden 2003). The generic version of the model is specified as follows: ΔY i,t = α + Δ X i,t β + Ф(Y i,t-1 - X i,t-1 Υ ) + є i,t, in which Y i,,t measures personnel spending in province i at time t and X is a vector of independent variables. The dependent variable is thus the annual change in personnel spending. The independent variables include the lagged level of personnel spending, the annual rate of change in the independent variables, and the lagged values of those independent variables. The underlying theoretical assumption is that the relationship between the variables in the model resembles a moving equilibrium in which the dependent variable may fluctuate in response to short-run changes in the independent variables but over the long run assumes levels consistent with those of the independent variables. The central advantage of the model is thus that it makes it possible to distinguish between the short and long-term relationship of X and Y. The former relationship is captured by β and the latter on the basis of the error correction term presented in parentheses. 15

18 Rewriting the generic model in the form in which it is actually estimated yields the following equation: ΔY i,t = β 0 + Y i,t-1 β 1 + ΔX i,t β 2 + X i,t-1 β 3 + ε i,t, in which β 1 is the same as Ф in the error correction version of the equation, β 2 is the same as β,and Фγ is rendered by β 3. The short-term relationship between changes in the independent variables and personnel spending is thus captured by β 2 and the longer term relationship by β 3, which is the coefficient of principal interest for hypotheses relating to issues such as government investment in economic development. Given problems of heteroskedasticity and contemporaneous correlation of errors across units that are common to cross-sectional time series research designs, the model is estimated on the basis of OLS with panel-corrected standard errors (PCSEs) (see Beck and Katz 1995). The results obtained via other methodologies are discussed below, together with alternative specifications of the base model involving the use of alternative measures of patronage spending, the inclusion of provincial dummy variables, and alternative control variables. Although the relative strength of the coefficients varies with the estimation technique, the variables measuring partisanship, electoral cycles, and fiscal policy choices remain relatively stable across different models and sets of estimations. Statistical Results Table 1 presents the statistical results for the model outlined above for the period as a whole. While there are some differences, the findings are relatively strong as well as consistent across the three indicators. Between 35 and 55 percent of the variance in patronage spending is explained, which is a reasonably high percentage given the nature of the statistical model. Unlike research focused upon the study of levels rather than 16

19 changes in spending, the value of the R 2 displayed at the base of the table is not inflated by the inclusion of a lagged dependent variable. With respect to the impact of fiscal policy, the data suggest that changes in the level of patronage spending over the long run reflect increases in provincial income, as indicated by the positive and statistically significant coefficients for lagged tax revenues, lagged transfers, and lagged nontax revenue. For the other two indicators of patronage, responsiveness to variations in revenue sources is weaker, although changes in transfers have significant short term effects on both per capita and percentage personnel spending. Lagged transfers also emerge as a long-term influence on changes in the relative priority placed on personnel spending. Also consistent with expectation is the tradeoff between patronage spending and capital investment, with the coefficient for lagged capital spending evincing a negative and statistically significant sign across all three sets of estimates. Drawing upon the error correction version of the model, these results can be readily expressed in terms of hypothetical changes in patronage allocations. Starting in an equilibrium state and holding all other variables constant, a 10 percent increase in tax income will be translated over the long run into a 7.8 percent increase in the level of personnel spending. Similarly, over the long run a 10 percent increase in transfer or nontax revenues national tax revenues will yield, respectively, a 7.7 and 7.5 percent increase in personnel spending. For a 10 percent increase in capital spending, on the other hand, the estimated impact is a 2.3 percent long-term decline in the level of personnel spending. As suggested by the coefficients in the second and third columns of Table 1, however, the effects of shifts in fiscal policy are much weaker for the per capita and percentage measures of personnel spending. 17

20 With respect to the other independent variables, the coefficients assessing unemployment are statistically significant but negative, suggesting that provinces suffering particularly severe economic setbacks over the past two decades have been forced to reduce their public sectors rather than compensate for economic insecurity via patronage spending. It may be estimated that a 1 percent increase in unemployment is translated over the long run into a 2.3 percent cut in the level of personnel spending. Adding estimates of provincial poverty levels to the three models presented in Table 1 does not alter this negative relationship. Unlike unemployment, however, insecurity generated by increased poverty has a positive and statistically significant impact on changes in both the level and per capita measures of personnel spending while lagged poverty significantly drives up the share of spending devoted to patronage (see Appendix I). 3 In terms of partisan effects, only presidential partisanship achieves statistical significance, raising questions about the importance that has been attributed to gubernatorial partisanship in the Argentine context. Party dominance, on the other hand, makes a difference, with two of the three coefficients suggesting that the narrower the electoral base of incumbent, the greater the incentives to spend on patronage rather than public goods. Replacing the indicator of party competition with a measure of the percentage difference between the vote for winning parties and their closest competitors yields similar but weaker results. Conceptualizing the relationship between competition and personnel spending as curvilinear does not alter this finding: an exponentiated term for party dominance does not achieve statistical significance within the framework of these three sets of estimations. Finally, with respect to the question of business cycles, the positive and statistically 18

21 significant coefficients for the post-gubernatorial election year, as distinct from the election or pre-election year, point to a pattern of post-electoral expansion of patronage spending in accordance with theoretical expectation. Reestimating the results separately for each of the three electoral dummy variables has no substantive effect on this finding: spending continues to rise significantly in the postelection year and the other electoral cycle dummies remain insignificant. Confidence in the meaning and significance of these findings is further enhanced by comparing the expenditure patterns of challengers with reelected incumbents. Due to high levels of intra- and inter-party competition as well as various types of term limits, only 25 percent of governors served consecutive terms between 1983 and It may be assumed that these governors were backed by unusually effective political machines: they were reelected with an average of 56.6 of the total vote as compared to 48.2 for other victorious candidates (differences that are statistically significant at the.001 level). The incentives for reelected governors to expand patronage in the wake of elections were therefore presumably reduced relative to other governors differences that should be reflected in personnel spending. Reestimating the three models presented in Table 1 with the addition of a dummy variable for reelected governors and an interaction term (reelected governor* postelectoral year) yields results that are consistent with this expectation. Whereas challengers significantly boosted personnel spending in the year following elections to fortify their electoral base, reelected governors tended to introduce cuts (see Appendix II). How robust are these findings regarding personnel spending? One way of answering this question is to alter the estimations presented in Table 1 by including provincial dummy variables to control for omitted variables that may determine long-term differences in 19

22 patronage spending. To summarize briefly, all of the coefficients that achieved statistical significance in two or more of the models presented in Table 1 remain significant, with the exception of that for party dominance, which achieves significance solely in the estimates for the share of personnel spending. The other major change is that with the addition of provincial dummy variables, the coefficient for presidential partisanship becomes consistently significant across the three estimates. The results are also relatively insensitive to changes in the case base. Restricting the analysis to the pre-2001 period, before intense crisis affected government spending, does not produce major changes in the results, except that the coefficients for presidential partisanship and lagged gubernatorial election achieve statistical significance at the.05 level in all three sets of estimations. Likewise, individual provinces can be deleted from the analysis without producing much change in the estimates. Deleting the most and least populous provinces from the estimates also has little impact, except with respect to the dummy variable for presidential partisanship, which again becomes more statistically significant. Alternative estimation techniques also point to the relative robustness of the statistical findings. OLS with robust standard errors clustered by province provides variance estimates that are less efficient although robust in the sense that they address panel-level heteroskedasticity as well as correlations among the observations of each panel. The resulting coefficients for transfers, capital expenditures, unemployment, partisanship, electoral year, and competitiveness are similar to those presented in Table 1, except stronger and consistently statistically significant in the case of the dummy variables for PJ President and lagged election year. Estimation utilizing random effects linear models likewise yield results that are relatively consistent with those reported above. 20

23 Arguably, however, the principal questions about the results reported in Table 1 are less ones of statistical robustness than of data validity. Since programmatic as well as patronage commitments are likely to be captured by personnel spending, do the available indictors adequately measure patronage? How distinctive is the responsiveness of patronage spending to variables such as the electoral cycle? Do more efficient forms of redistributive spending follow different patterns? To address these questions, Table 2 below draws upon official data for the period that disaggregate spending into functional categories, separating out central administration expenditures from personnel expenditures associated with the delivery of services, including social spending (Ministerio de Economía y Producción, a). The administrative category is limited to the share of spending absorbed by the central administration of the province--expenditures that may more closely approximate the concept of patronage than total personnel spending. Social spending is measured as the share of spending devoted to health, education, welfare, social security, housing, and related services. The third category of expenditure included in the estimates is comprised of goods and services significant for economic growth, including spending on physical infrastructure. While social and economic spending may well include transfers to targeted groups of voters, presumably both categories also involve significant expenditures on public and quasi-public goods. Taken together the three categories of expenditure account for an average of 88.6 percent of total provincial spending during the time period under analysis, with the remainder consisting mainly of spending on police and internal security. If patronage spending represents an inefficient form of social redistribution, administrative and social spending might not respond very differently to economic 21

24 insecurity or to the budgetary priority accorded to economic development. Building on the theoretical arguments elaborated above, however, administrative spending is predicted to evince a distinctive response to nontax revenues, partisanship, and the electoral cycle. Nontax revenues (including federal transfers) are expected to foster bureaucratic expansion, not increased expenditure on public goods. PJ control of the presidency or governorship is likewise expected to be translated into increased patronage rather than social spending. The impact of the electoral cycle on social and patronage spending is also expected to generate divergent results. In accordance with the logic of the political business cycle, social spending is expected to expand in advance of elections and patronage spending in their wake. Finally, to the extent that electoral fragmentation erodes incentives for the delivery of public goods, low levels of electoral support might be expected to be reflected positively in administrative spending but negatively in social spending. Drawing upon seemingly unrelated regression (SUR), Table 2 explores these expectations on the basis of two models of provincial budgetary priorities. The first model conforms to that presented in Table 1 above. The second model is identical, except for the addition of an exponentiated term to assess differences in the responsiveness of administrative and social spending to political competition. [INSERT TABLE 2 ABOUT HERE] The findings of the analysis are consistent with expectation as well as reasonably compelling with respect to the distinctiveness of patronage and social spending. With respect to revenue sources, the estimates indicate that administrative spending rises with government transfers, whereas social spending rises in response to higher tax revenues. The difference suggests that the tighter the nexus between between spending and taxation the 22

25 less the propensity for patronage politics. Similar differences between social and administrative spending emerge with respect to the impact of unemployment, partisanship, and the electoral cycle. Whereas social spending responds to short term shifts in unemployment, administrative spending increases with unemployment over the long run, offering support for the linkage that has been made between patronage politics and economic insecurity. Partisanship also affects administrative and social spending differently, with Partido Justicialista control of the presidency strongly associated with increases in administrative, but not social, spending. The electoral cycle also shapes spending patterns in accordance with expectation, with social spending increasing before and during gubernatorial elections and declining in their wake. The negative coefficient for administrative spending during election years, on the other hand, points to the opposite pattern. 4 Although the results are weak and heavily influenced by particular observations, the contrasts also hold up in terms of political competition, with the coefficients reported for Model 2 suggesting that the relationship between political competition and patronage spending assumes a U shape, whereas that for competition and social spending approximates an inverted U. Hence as political support for the gubernatorial party moves from low to medium levels, spending on central administrative personnel declines; but as gubernatorial support shifts from medium to high levels, such spending increases. The contrasts between social and administrative spending break down mainly in relationship to economic policy. The greater the priority accorded to spending on economic development, the lower spending on both administration and social services. Drawing upon the Breusch-Pagan test of independence, joint estimation reveals that social and patronage 23

26 spending are interdependent and negatively related. In both sets of estimations, the error terms for administrative and social spending are correlated at the.001 level. The same is not true of the relationship between administrative and police spending, which is to say that the tradeoffs are not an artifact of the focus on spending shares. It may also be noted that reestimating these results in OLS with panel-corrected standard errors yields similar findings. Accordingly, focusing narrowly on administrative as distinct from total personnel spending not only emphasizes the contrasting incentives for patronage and social spending but also generates results that are consistent with theoretical expectation. Finally, with respect to what is perhaps the most distinctive claim of the preceding analysis namely, the existence of patronage business cycles, Table 3 offers a simple oneway analysis of variance of spending patterns, which is designed to emphasize that the preceding results are not simply the product of a particular statistical model or estimation technique. Consistent with the evidence presented in Table 2, the mean level of spending on central administrative personnel rises after elections, whereas social spending rises in advance. The bivariate analyses for changes in personnel spending, personnel spending per capita, and personnel spending as a share of the total budget yield equivalent results. [INSERT TABLE 3 ABOUT HERE] Conclusion The literature on the political economy of democracy takes programmatic linkages between citizens and politicians as its point of theoretical departure. Consistent with this perspective, partisan, political business cycle, and rational expectations models of economic policy performance all presume that the electorate responds to the prospective policy commitments and/or past performance of politicians. While the resulting research program has proved very fruitful, it is nevertheless limited by its restrictive conception of electoral 24

27 competition and narrow empirical base. With few exceptions, scholars have generalized about the relationship between elections and economic policy on the basis of data on advanced industrial nations, largely ignoring the pervasiveness of patronage politics in other regions of the world. Bringing together insights from research on clientelism and the political economy of elections, this study advances our understanding of both phenomena by developing and testing a model of patronage spending that emphasizes the strategic choices of politicians. Drawing upon cross-sectional time series evidence on subnational politics in Argentina, the study offers some support for the traditional literature on patronage politics, which tends to stress the importance of poverty and insecurity to the development and maintenance of political machines. These factors, however, play only a weak and inconsistent role in accounting for variations in patronage spending in Argentina. The evidence more strongly highlights the importance of variables that figure prominently in the literature on the political economy of democracy. Within the common framework established by Argentine national institutions and political culture, the partisanship of elected officials, stage of the electoral cycle, pattern of revenue generation, and spending on economic development give rise to significant variations in levels of patronage spending. Important and largely neglected parallels between patronage and programmatic politics thus emerge from the analysis. Given partisan differences in the credibility of political commitments, partisanship shapes policy choice and performance under both patterns of political competition. Likewise, political business cycles emerge in response to both patronage and programmatic politics, with a propensity for the expansion of patronage 25

28 emerging in the wake of elections. Although the evidence is weak and not fully consistent across the two major components of the preceding analysis, variations in patronage spending also appear to reflect variations in political competition, with governors supported by relative narrow partisan bases relying more heavily upon public sector employment than their counterparts. Thus instead of explaining patronage politics solely or even mainly in terms of relatively immutable contextual conditions, such as poverty or political culture, this study emphasizes the benefits of exploring the relationship of patronage spending to variables of broader relevance for the study of the political economy of elections. Also of importance for understanding variations in patronage spending are the fiscal policy choices of public officials. The findings of the preceding analysis suggest that the structure of taxation and revenue generation significantly influences the level of personnel spending. The most consistent finding involves the responsiveness of patronage to discretionary intergovernmental transfers, suggesting that weak linkages between taxing and spending decisions enhance the tendency toward patronage spending. The evidence also reinforces recent theoretical work suggesting that politicians relying on patronage networks will seek to reinforce their positions by choosing policies that privilege inefficient redistribution via targeted benefits over socioeconomic development. In the Argentine provinces, patronage spending is negatively related to spending on economic development as well as to spending on social services. In addition to underlining the relevance of the literature on the political economy of democracy to the study of patronage politics, this study also suggests ways in which the study of patronage politics may enhance our understanding of political economy. Particularly notable is the evidence regarding the relationship of patronage to the electoral 26